计算机工程与应用2026，Vol.62Issue(17)：295-305,11.DOI:10.3778/j.issn.1002-8331.2507-0169
融合语义校正与背景抑制的多光谱目标检测方法
Multispectral Object Detection Method Integrating Semantic Correction and Background Suppression
摘要
Abstract
Multispectral object detection suffers from semantic conflicts and background-induced false positives due to modality heterogeneity and complex scenes.This paper proposes the semantic dual collaborative optimization network(SDCONet),which adopts an align-then-enhance paradigm for multi-scale,end-to-end collaborative optimization.To resolve semantic conflicts,it introduces cross-modal semantic correction(CSC),which explicitly measures global cross-modal similarity via normalized correlation maps between RGB and infrared(IR)features.CSC combines k-nearest neighbors(KNN)scheme to select semantically similar fine-grained neighbors for aggregation,enabling dynamic align-ment that couples global dependency capture with local semantic matching,unlike static attention based approaches,mitigates cross-modal misalignment at its source.To curb background false positives,the paper proposes semantic-background con-trastive augmentation(SBCA),which employs a global channel communication(GlobalCC)mechanism to establish cross-modal,channel-wise semantic associations and combines foreground-mask supervision with foreground/background con-trastive learning to strengthen foreground representations while suppressing background noise,overcoming the limitations of unimodal channel optimization.Experiments on the FLIR and KAIST multispectral datasets show that SDCONet improves mAP by 2 and 3 percentage points over the ICAFusion baseline,respectively,while reducing FPR by an abso-lute 1 percentage point on both datasets,demonstrating superior detection accuracy and robustness.关键词
多光谱目标检测/跨模态语义校正(CSC)/语义背景对比增强(SBCA)/特征融合Key words
multispectral object detection/cross-modal semantic correction(CSC)/semantic-background contrastive aug-mentation(SBCA)/feature fusion分类
信息技术与安全科学引用本文复制引用
谢斌红,任良,张睿,张英俊..融合语义校正与背景抑制的多光谱目标检测方法[J].计算机工程与应用,2026,62(17):295-305,11.基金项目
吕梁市引进高层次科技人才重点研发项目(2022RC08) （2022RC08）
山西省产教融合研究生联合培养示范基地项目(2022JD11) （2022JD11）
山西省科技成果转化引导专项项目(202304021301020). （202304021301020）
